-
5f30fc7031
[Bugfix][KV Connector][Mooncake] Suppress completion for empty pulls (#59347)
Yicong Wang
2026-10-03 15:28:00 +08:00
-
b0e21b3083
[ROCm][Kimi-K3] Add opt-in gfx942 MXFP4-to-int4 conversion (#51274)
Markus Hartikainen
2026-10-02 23:28:02 -07:00
-
ca0df1f122
[Bugfix][Core] Exempt exactly the blocks an async KV load writes from zeroing (#59504)
2026-10-02 23:26:12 -07:00
-
1a001d5842
[CI] Drop duplicate bf16 skinny GEMM test from Kimi K3 B200 job (#59850)
Kevin H. LuuandClaude Opus 5.5
2026-10-02 21:43:05 -07:00
-
c28834df00
[ROCm][CI] Extend AMD coverage for distributed, model, and eval tests (#56679)
2026-10-02 23:20:51 -05:00
-
28556f8600
[Bugfix] Log CRIU failure details before snapshot cleanup (#59661)
Nils Matteson
2026-10-02 21:02:27 -07:00
-
9fdb174678
[Bugfix] Avoid InfiniBand state in TP1 snapshots (#59699)
Nils Matteson
2026-10-02 20:59:17 -07:00
-
faa9860dbd
[ROCm][CI][AiterExperts] Add test coverage for hidden/intermediate padding correctness (#59333)
Divakar Verma
2026-10-02 21:53:37 -04:00
-
44198f577f
[Bugfix][Model] Fix M-RoPE offset double-count in Qwen3-Omni (#58890)
C
2026-10-03 09:00:51 +08:00
-
8c3b720a4d
[TEST][XPU][CI] disable xpu tests for nonexistent input norm kernels (#59568)
Miłosz GrunwaldandKunshang Ji
2026-10-03 02:27:29 +02:00
-
7dfe3338d5
[EPLB] Add contention-aware expert migration batching (#52641)
DOCCA0andRobert Shaw
2026-10-03 08:13:50 +08:00
-
8e443871f0
[CI/Build][NVIDIA] Build Rubin images on the public nvidia/cuda base image (#59288)
2026-10-02 20:12:33 -04:00
-
8bdd8d847d
[GLM 5.3 Perf] Enable fused multi-step decode, 13.3% E2E throughput improvement for concurrency 1 (#57443)
Wentao Ye
2026-10-02 20:05:27 -04:00
-
6e517b15c1
[Docs] Add ERNIE 4.5 to batch invariance tested models (#58476)
2026-10-02 13:37:32 -07:00
-
84593955fa
[Perf][MoE] Support fp8 combine in FlashInfer one-sided MoE all2all (#57995)
Ayaan Nadamal
2026-10-03 00:29:16 +04:00
-
111f71a6db
[feat] FlashInfer CuteDSL MegaMoE integration (#54049)
Julien Debache
2026-10-02 21:43:33 +02:00
-
c4973d94fe
[Bugfix][Frontend] Avoid generation for empty streaming input (#59015)
Benedikt Falk
2026-10-02 20:48:43 +02:00
-
10f6ab01a9
[Perf] Use value-only reduction for native per-token FP8 quantization (#59800)
jackLeiandMisha Goin
2026-10-02 11:37:17 -07:00
-
8882b83b34
[Bugfix] Bump tokenizers to 0.23.2 for duplicate-pattern support (#59796)
djramic
2026-10-02 20:33:50 +02:00
-
202d497aa8
[Bugfix][Watermarking] Keep draft prompt lengths valid under CUDA graphs (#59779)
simon-veitner-redhatandLucas Wilkinson
2026-10-02 20:32:11 +02:00
-
d1a974c140
[Frontend] Constrain non-strict GLM-4.7 tool calls with a shallow structural tag (#56403)
Shijin ZhangandFlora Feng
2026-10-03 02:31:53 +08:00
-
f700178f90
[Refactor] Remove dead env and config (#59781)
Wentao Ye
2026-10-02 14:07:20 -04:00
-
4e4d75c459
[GLM5.3 Perf] Reuse sparse MLA index conversion across layers, 3.5~3.9x kernel performance improvement (#59464)
Wentao Ye
2026-10-02 13:37:45 -04:00
-
7cfd233538
Revert "[Bugfix] Tie lm_head.weight for Nemotron Parse when checkpoint omits it" (#53020) (#59805)
Thang NguyenandClaude Opus 5.5
2026-10-02 13:19:25 -04:00
-
f590eb2448
[Perf][Qwen4Exp] Add SM121 TP=1 skinny-GEMM plans (#59753)
Stefano CastagnettaandBenjamin Chislett
2026-10-02 19:17:45 +02:00
-
9cf087fd14
[Perf] Tune MoE weighted-sum kernel launch configuration (#59731)
2026-10-03 01:09:29 +08:00
-
8442117446
Upgrade tpu-inference to v0.30.0 (#59396)
2026-10-02 16:39:03 +00:00
-
1ab18599ca
[Agents] Add pre-commit check for agent files and skills (#59657)
Misha GoinandClaude
2026-10-02 12:38:22 -04:00
-
e42d35d4f3
[Minimax-M3] Keep the native FP8 MMA in the Triton indexer scorers (#59481)
rmhaskarandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-10-02 09:30:39 -07:00
-
167a81fa9c
[CI] Fix ModelExpress handling in weight transfer tests (#59772)
djramic
2026-10-02 18:30:01 +02:00
-
6abadc2aa7
[ROCm][CI] Add test coverage for VLLM_ROCM_MOE_PADDING memory-stride padding transparency (#59332)
Divakar Verma
2026-10-02 12:23:22 -04:00
-
2a537887de
[CI] Use vllm_runner in fusions_e2e conftest for reliable GPU cleanup (#59525)
Divakar Verma
2026-10-02 12:21:21 -04:00
-
7b6be2f1f4
[Docs] Add Reviewers page (#59459)
+8
2026-10-02 12:11:31 -04:00
-
4c4003bb5f
[Bugfix] Bind routed-experts capture to the MoE layer, not the kernel (#59455)
2026-10-02 08:50:38 -07:00
-
0979892992
[Bugfix][Multimodal] Fix GLM-5.3-Flash vision tower crashes on image input (#59126)
2026-10-02 11:25:37 -04:00
-
f0c44cc7d5
[Bugfix] Fix stalled local-only Elastic EP scale-up (#57820)
Taneem Ibrahim
2026-10-02 09:47:55 -05:00
-
592c6f3fbc
[Bugfix][Pooling] Handle multimodal cache misses without crashing (#58975)
2026-10-02 22:14:33 +08:00
-
443dccd3c9
[Bugfix][Multimodal] Fix reference counting of the SHM processor cache (#59695)
tenotekandClaude Fable 5.1
2026-10-02 22:14:13 +08:00
-
c91dccc080
[Bugfix][Quark] Pass grouped-routing arguments to OCP MX monolithic kernels (#59752)
aoshen02andClaude Opus 5.5
2026-10-02 21:46:48 +08:00
-
e4294144a8
[Perf] Keep non-speculative GDN decode on the standard path (#59735)
2026-10-02 22:44:53 +09:00
-
24df2d3fc7
[Model] Add Qwen4Exp to the Qwen GDN Triton warmup (#56742)
2026-10-02 22:38:15 +09:00
-
30e956fad3
[Docs] Fix stale W4A16 NVFP4 default kernel in ModelOpt docs (#59135)
Rajeev Warrierandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-10-02 18:51:43 +05:30
-
d648847fe4
[CI][Kimi-K3] Test prefix cache reuse with KV offload, P/D and DCP (#59229)
Jiangyun ZhuandClaude Opus 5.5
2026-10-02 20:58:35 +08:00
-
f0b4b36890
[Bugfix][Quantization] Detect NVFP4 in ModelOpt mixed-precision checkpoints (#56050)
Stefano CastagnettaandClaude Opus 5
2026-10-02 14:30:22 +02:00
-
fa212b8dc5
[Test] Re-enable Ovis2.5 and Ovis2.6-MoE vLLM tests on Transformers v5 (#59763)
Abhay ShuklaandClaude Opus 5.5
2026-10-02 17:29:40 +05:30
-
bdfb02674f
[XPU][Kernel] Tune Triton W8A8 block-FP8 GEMM for Intel B70 (#56063)
pmanczakandPawel Manczak
2026-10-02 13:56:11 +02:00
-
e2cb5ee46c
[CI] Replace shellcheck-suppressed patterns flagged in #52572 with clean equivalents (#53341)
Turner JabbourandClaude Sonnet 5
2026-10-02 05:30:04 -06:00
-
6e4efafc6e
[Model] Migrate GPT-NeoX, Phi, Seed-OSS and Jais2 to the Transformers modeling backend (#59701)
Harry MellorandClaude Opus 5.5
2026-10-02 12:18:35 +01:00
-
5d20979fc9
[Misc] Remove code paths for Transformers < 5.16.1 (#59762)
Harry MellorandClaude Opus 5.5
2026-10-02 12:17:24 +01:00
-
8c60714da5
[Bugfix][LoRA] Fall back for high-rank MoE LoRA (#55161)
2026-10-02 18:59:28 +08:00
-
98b29aa99a
[ROCm][Perf][GLM-5.3-Flash] Add AITER topk backend for decodes (#58167)
Simon Danielssonandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-10-02 12:47:41 +02:00
-
751fa96a49
[ROCm][Perf][GLM-5.3-Flash] Remove redundant copy after ragged sparse MLA (#58569)
Simon Danielssonandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-10-02 12:45:51 +02:00
-
5688a4dd4a
[Bugfix][GLM-5.3] Size the image encoder cache from the exact token ceiling (#59565)
2026-10-02 18:43:29 +08:00
-
af2a582d5c
[Bugfix][Spec Decode] Qualify Transformers backend attention layer names with the model prefix (#59293)
2026-10-02 16:21:20 +06:00
-
8e58ac22af
[Feature] Offload KV-init runtime state on sleep (#59158)
aoshen02andClaude Opus 5.5
2026-10-02 17:49:21 +08:00
-
c15672cfb4
[Security] Accept zero-sum DeepSeek-OCR pixel tensors (#59417)
Juan Pérez de Algaba
2026-10-02 11:17:12 +02:00
-
-
58b3298457
[Misc] Add Transformers version upper bound in requirements (#59614)
Isotr0pyandHarry Mellor
2026-10-02 16:27:11 +08:00
-
9d2a6f52b2
[Rust Frontend] Expose gRPC port in Python vllm serve (#59659)
2026-10-02 01:14:37 -07:00
-
b558f160a2
[Bugfix] Fix minimax-m3 multimodal processor compatability with Transformers v5.18 (#59613)
Isotr0py
2026-10-02 15:29:49 +08:00
-
0154979656
[ROCm][Perf] Kimi-K3 enable prefill checkpoints on ROCm (#58344)
kliuae
2026-10-02 14:58:04 +08:00
-
4056c8ac1f
[Bugfix][HiSparse] Fix MTP acceptance collapse under FULL graphs with a saturated GPU pool (#59309)
2026-10-02 02:04:21 -04:00
-
0cbac6cd13
[Model][Spec Decode] Enable EAGLE3/DSpark pipeline parallelism for Sarvam MLA (#55902)
2026-10-02 09:58:19 +05:30
-
f7999d2e44
[Bugfix][HiSparse] Size the KV cache from the groups HiSparse allocates (#59450)
2026-10-02 00:11:55 -04:00
-
1a9eaa3a68
[Doc] Document ECMooncakeConnector for EPD disaggregation (#59196)
Teng MaandTeng Ma
2026-10-02 12:10:33 +08:00
-
8486f1ebfe
[Perf][PCP] Shard decode requests across PCP ranks (#52162)
Qiu ChunshuoandLucas Wilkinson
2026-10-02 09:20:13 +05:30
-
a810316c08
[Bugfix] Keep batch-invariance NCCL pins out of the weight-transfer group (#59500)
2026-10-01 20:19:28 -07:00
-
fcfacc3499
[Bugfix][Frontend] Seed derender detokenization from the prompt (#59046)
Martin Hickey
2026-10-02 04:17:56 +01:00
-
8faacd707f
[Model] Migrate Glm, Arcee, CWM and Mellum to the Transformers modeling backend (#59679)
Harry MellorandClaude Opus 5.5
2026-10-02 04:15:14 +01:00
-
ea106db806
[Bugfix][Bench] Clean up synthetic video files and writers (#59529)
Prudhvi VudaandOpenAI Codex
2026-10-01 22:51:34 -04:00
-
9bd993564f
[Frontend] Parse tool calls and reasoning from checkpoint response templates (#58604)
2026-10-01 22:33:00 -04:00
-
5a661487e2
[Feature] Add native ModelExpress weight transfer backend (#58399)
Hyunjae Wooandaoshen02
2026-10-01 19:29:35 -07:00
-
785b990116
[Distributed] Enable custom all-reduce under VLLM_BATCH_INVARIANT (#58623)
Flora Feng
2026-10-01 22:25:48 -04:00
-
e3cae8d2ac
[Model] Use upstream GLM-5.3 and Qwen4-Exp configs and processor (#57387)
Harry MellorandClaude Opus 5
2026-10-02 03:25:18 +01:00
-
866f02db8a
[Feature] Release WorkspaceManager scratch on sleep (#59156)
aoshen02andClaude Opus 5.5
2026-10-02 10:14:37 +08:00
-
e5d33a36e7
[Bugfix][Rust Frontend] Preserve whitespace in GLM string arguments (#59654)
2026-10-01 19:10:35 -07:00
-
6660926d92
[Bugfix][CPU] Fix macOS build on Apple Clang < 17: structured binding… (#58932)
2026-10-01 19:01:07 -07:00
-
4caa060053
[TEST][CI] Fix serve rlhf tests subprocess import errors (#59622)
2026-10-02 03:47:47 +02:00
-
89772ddb1f
[CI][XPU] Skip CUDA-IPC weight sync metrics test on Intel (#59556)
liuzhenwei
2026-10-02 09:28:58 +08:00
-
d58da79de7
[Metrics][P/D] Add KV-fetch stage gauges for async KV loads (#58874)
2026-10-01 18:24:33 -07:00
-
eec3c36a47
[ROCm][BugFix] Revert AITER PA gluon decode from ROCM_AITER_FA (#54805)
2026-10-01 17:21:18 -07:00
-
5b6b657e13
[ci] Update mergify to rebase if behind by 100 commits (#59710)
Kevin H. Luu
2026-10-01 17:17:37 -07:00
-
28c57456db
[Bugfix][Frontend] Accept Anthropic tool_addition and tool_removal content blocks (#57693)
2026-10-02 05:33:36 +05:30
-
e6674262f1
[Bugfix][CI] Widen DBO+DP+EP GSM8K accuracy margin on ROCm (#59700)
Divakar VermaandClaude Sonnet 5
2026-10-01 19:37:36 -04:00
-
e35082e7b6
[Bugfix][CLI] Include inherited field docstrings in get_attr_docs (#49821)
hclandHarry Mellor
2026-10-02 07:35:58 +08:00
-
d2125ea681
[ROCm][CI] Raise the MI355 DeepSeek-R1 GSM8K startup wait to 1800s (#59666)
Aarushi Jain
2026-10-01 17:30:13 -05:00
-
a5105dba05
[Frontend] Port Step-3.5 parsers to the streaming parser engine (#59321)
Flora Feng
2026-10-01 18:18:37 -04:00
-
ca65eb67d9
[Perf][HiSparse] Avoid repeated prefix scans and residency updates (#57930)
2026-10-01 14:48:18 -07:00
-
7566d83bd3
[Attention][MiniMax-M3] NVFP4 KV cache on the MSA sparse attention path (#59300)
Yongye ZhuandClaude Opus 5.5
2026-10-01 14:39:55 -07:00
-
ea84051819
[KVConnector][NIXL] Count completion notifications that arrive after KV expiry (#58875)
Zijing Liu
2026-10-01 14:30:58 -07:00
-
ccda9098cc
[ROCm][CI] Drop two no-GPU AMD mirrors from the CPU test areas (#59593)
stefankoncarevicandAndreas Karatzas
2026-10-01 21:54:26 +02:00
-
840c7c2f4b
[Bugfix][Engram] Fix intermittent Triton 3.8 crash in the lookup kernel (#59639)
2026-10-01 21:38:55 +02:00
-
7b44aad5db
[MyPy] Fix mypy errors in
vllm/model_executor/models/[kK]* (#59402)
Harry MellorandClaude Opus 5.5
2026-10-01 20:32:27 +01:00
-
2a52715ef7
[Bugfix][Responses] Use standard reasoning content-part events (#59652)
2026-10-01 12:20:57 -07:00
-
d848c4ed3d
[ROCm][Triton] Migrate Kimi-K3 kernels from make_block_ptr to tensor … (#58769)
Jaden Mathias
2026-10-01 14:41:41 -04:00
-
47de9d4d04
[KV Connector][NIXL] Coalesce host-buffer KV copies across cache groups (#54483)
Chanbin LimandChendi.Xue
2026-10-01 11:10:54 -07:00
-
e9be5323f0
[Perf] Integrate flash-maxsim Triton kernels for late-interaction scoring (#40337)
2026-10-01 21:06:43 +03:00
-
083060d046
[CI/Build] Add agents auto-label rule (#59641)
Misha GoinandClaude
2026-10-01 13:28:27 -04:00
-
3a69636645
[Bugfix][HiSparse] Fix a chunked-prefill preemption livelock (#59494)
Matthew BonanniandClaude Opus 5.5
2026-10-01 13:02:31 -04:00
-
26bfdfb6bc
[ROCm][Perf] Parallelise AITER MLA page-index expansion over token chunks (#57978)
2026-10-01 10:00:15 -07:00
-
362cfb72de
[Agents] Expose PR checklist skill to Claude (#59638)
Misha GoinandClaude
2026-10-01 12:54:54 -04:00